[Tagging] Verifiability wiki page: "Geometry" section added

[Christoph Hormann]

On Sunday 28 April 2019, Tobias Knerr wrote:
>
> Yes, it's often not possible to agree on a precise border for these
> features. But nevertheless, there are typically areas that are
> definitely part of them, and other areas there are definitely not
> part of them.

I'd like to emphasize once more that verifiability has absolutely 
nothing to do with mapping precision or measurement errors.

It is the nature of every geographic feature (in the sense of something 
with at least limited localizability) that you can specify locations 
that are definitely not part of it.  That does not mean you can create 
a verifiable polygon geometry for them.

OTOH just because the only information source you have for example about 
a lake is a poor quality GPS track from a walk around it does not mean 
you cannot map that lake with a polygon based on that information.  
Because the problem here is not your limited ability to localize the 
shore of the lake but the ability to precisely measure it.

Getting this difference is key to understanding the essence of 
verifiability on OSM.

> The above is verifiable geographic information, so it ought not be
> off-limits for OSM.

So you think "The Amazon Rainforest" should be mappable with a polygon 
in OSM because you can make a verifiable statement that it does not 
extend to Asia?

> [...] One could
> imagine, for example, a relation containing two polygons for the
> feature's "minimum" and "maximum" extent (describing the parts of the
> world that are verifiably part of/not part of the feature), with a
> grey area of uncertainty in between.

Seriously?

Because one polygon is not a verifiable representation of a certain 
feature you want to replace it with - drumroll - two polygons?

The idea of developing new data objects for selectively documenting 
verifiable information of objects with limited localizability is not 
necessarily bad but the aim of this would need to be to allow limiting 
the recorded information to exactly what can verifiably be said about a 
feature, not to add more non-verifiable data to disguise the 
non-verifiable nature of the whole thing.  I have thought about this 
quite a bit and came to the conclusion that the answer to this is 
usually that using a node rarely misses any verifiable information that 
cannot most efficiently be recorded in the form of tags or that is not 
already recorded implicitly in the form of other features that are on 
their own much better verifiable.

> With your recommended solution of placing a node "near the center of
> the feature", capturing this useful knowledge is not possible. It
> also doesn't make logical sense to me: If it were indeed impossible
> to verifiably establish even an approximate boundary of the feature,
> how can we verifiably establish the feature's center?

I have had this discussion plenty of times in the past - the 
verifiability of a node localizing a feature is much easier to achieve 
(through a clear rule where to place the node) and demonstrate than for 
a polygon.  Verifiability of a node location means nothing more than 
that qualified independent placements of the node converge to a single 
location.  This is a completely scale independent condition - it can be 
fullfilled with a standard derivation of less than a meter (for 
something like a power=pole) but might also be many kilometers.  For a 
polygon that is not the case.

I don't really like the extension Joseph wrote on the Verifiability page 
but not because i disagree with the general idea but because for my 
taste it is too much *definition by example* which is a poor way of 
communicating the concept in general.  Examples are useful and needed 
to clarify the meaning (and they have been used as such on that page 
for a long time) but they are no replacement for formulating the 
general abstract idea behind verifiability in a compact form that is 
not tied to specific examples.  Andy's idea of creating subpages 
explaining how to practically apply verifiability to tags and 
geometries is probably the right approach.

-- 
Christoph Hormann
http://www.imagico.de/